Merge branch 'feat/custom_schema_and_layout' of github.com:presenton/presenton into feat/custom_schema_and_layout

servers/fastapi/api/v1/ppt/endpoints/images.py

@@ -1,4 +1,3 @@
-from typing import Annotated
 from fastapi import APIRouter, Body
 
 from models.image_prompt import ImagePrompt

servers/fastapi/get_test_schema.py

@@ -2,6 +2,7 @@ from typing import List, Optional
 from pydantic import BaseModel, Field, HttpUrl, EmailStr
 
 from models.presentation_layout import PresentationLayoutModel, SlideLayoutModel
+from models.presentation_outline_model import PresentationOutlineModel
 
 
 class ContactInfoModel(BaseModel):
@@ -559,4 +560,6 @@ presentation_layout = PresentationLayoutModel(
     ],
 )
 
-print(presentation_layout.model_dump_json())
+# print(presentation_layout.model_dump_json())
+
+print(PresentationOutlineModel.model_json_schema())

servers/fastapi/utils/async_iterator.py

@@ -0,0 +1,16 @@
+import asyncio
+from typing import AsyncGenerator, Callable, Iterator, TypeVar
+
+T = TypeVar("T")
+
+
+def iterator_to_async(
+    func: Callable[..., Iterator[T]],
+) -> Callable[..., AsyncGenerator[T, None]]:
+    async def wrapper(*args, **kwargs) -> AsyncGenerator[T, None]:
+        iterator = func(*args, **kwargs)
+        for item in iterator:
+            yield item
+            await asyncio.sleep(0)
+
+    return wrapper

servers/fastapi/utils/llm_calls/generate_presentation_outlines.py

@@ -1,10 +1,14 @@
 from typing import Optional
 from openai.lib.streaming.chat._events import ContentDeltaEvent
+from google.genai.types import GenerateContentConfig
 
+from utils.async_iterator import iterator_to_async
 from utils.get_dynamic_models import get_presentation_outline_model_with_n_slides
 from utils.llm_provider import (
+    get_google_llm_client,
     get_large_model,
     get_llm_client,
+    is_google_selected,
 )
 
 system_prompt = """
@@ -64,18 +68,33 @@ async def generate_ppt_outline(
     model = get_large_model()
     response_model = get_presentation_outline_model_with_n_slides(n_slides)
 
-    client = get_llm_client()
-    async with client.beta.chat.completions.stream(
-        model=model,
-        messages=get_prompt_template(prompt, n_slides, language, content),
-        response_format={
-            "type": "json_schema",
-            "json_schema": {
-                "name": "PresentationOutline",
-                "schema": response_model.model_json_schema(),
+    if not is_google_selected():
+        client = get_llm_client()
+        async with client.beta.chat.completions.stream(
+            model=model,
+            messages=get_prompt_template(prompt, n_slides, language, content),
+            response_format={
+                "type": "json_schema",
+                "json_schema": {
+                    "name": "PresentationOutline",
+                    "schema": response_model.model_json_schema(),
+                },
             },
-        },
-    ) as stream:
-        async for event in stream:
-            if isinstance(event, ContentDeltaEvent):
-                yield event.delta
+        ) as stream:
+            async for event in stream:
+                if isinstance(event, ContentDeltaEvent):
+                    yield event.delta
+
+    else:
+        client = get_google_llm_client()
+        generate_stream = iterator_to_async(client.models.generate_content_stream)
+        async for event in generate_stream(
+            model=model,
+            contents=[get_user_prompt(prompt, n_slides, language, content)],
+            config=GenerateContentConfig(
+                system_instruction=system_prompt,
+                response_mime_type="application/json",
+                response_json_schema=response_model.model_json_schema(),
+            ),
+        ):
+            yield event.text

servers/fastapi/utils/llm_calls/generate_slide_content.py

@@ -1,36 +1,48 @@
+import asyncio
 import json
-from pydantic import BaseModel, Field
+from google.genai.types import GenerateContentConfig
 from models.presentation_layout import SlideLayoutModel
 from models.presentation_outline_model import SlideOutlineModel
-from utils.llm_provider import get_llm_client, get_small_model
+from utils.llm_provider import (
+    get_google_llm_client,
+    get_llm_client,
+    get_small_model,
+    is_google_selected,
+)
+
+system_prompt = """
+    Generate structured slide based on provided title and outline, follow mentioned steps and notes and provide structured output.
+
+    # Steps
+    1. Analyze the outline and title.
+    2. Generate structured slide based on the outline and title.
+
+    # Notes
+    - Slide body should not use words like "This slide", "This presentation".
+    - Rephrase the slide body to make it flow naturally.
+    - Do not use markdown formatting in slide body.
+"""
+
+
+def get_user_prompt(title: str, outline: str):
+    return f"""
+        ## Slide Title
+        {title}
+
+        ## Slide Outline
+        {outline}
+    """
 
 
 def get_prompt_to_generate_slide_content(title: str, outline: str):
     return [
         {
             "role": "system",
-            "content": f"""
-                Generate structured slide based on provided title and outline, follow mentioned steps and notes and provide structured output.
-
-                # Steps
-                1. Analyze the outline and title.
-                2. Generate structured slide based on the outline and title.
-
-                # Notes
-                - Slide body should not use words like "This slide", "This presentation".
-                - Rephrase the slide body to make it flow naturally.
-                - Do not use markdown formatting in slide body.
-            """,
+            "content": system_prompt,
         },
         {
             "role": "user",
-            "content": f"""
-                ## Slide Title
-                {title}
-
-                ## Slide Outline
-                {outline}
-            """,
+            "content": get_user_prompt(title, outline),
         },
     ]
 
@@ -38,24 +50,37 @@ def get_prompt_to_generate_slide_content(title: str, outline: str):
 async def get_slide_content_from_type_and_outline(
     slide_layout: SlideLayoutModel, outline: SlideOutlineModel
 ):
-    response_format = {
-        "type": "json_schema",
-        "json_schema": {
-            "name": slide_layout.name or slide_layout.id,
-            "schema": slide_layout.json_schema,
-        },
-    }
 
-    client = get_llm_client()
     model = get_small_model()
 
-    response = await client.beta.chat.completions.parse(
-        model=model,
-        messages=get_prompt_to_generate_slide_content(
-            outline.title,
-            outline.body,
-        ),
-        response_format=response_format,
-    )
-
-    return json.loads(response.choices[0].message.content)
+    if not is_google_selected():
+        client = get_llm_client()
+        response = await client.beta.chat.completions.parse(
+            model=model,
+            messages=get_prompt_to_generate_slide_content(
+                outline.title,
+                outline.body,
+            ),
+            response_format={
+                "type": "json_schema",
+                "json_schema": {
+                    "name": "SlideContent",
+                    "schema": slide_layout.json_schema,
+                },
+            },
+        )
+        return json.loads(response.choices[0].message.content)
+    else:
+        client = get_google_llm_client()
+        user_prompt = get_user_prompt(outline.title, outline.body)
+        response = await asyncio.to_thread(
+            client.models.generate_content,
+            model=model,
+            contents=[user_prompt],
+            config=GenerateContentConfig(
+                system_instruction=system_prompt,
+                response_mime_type="application/json",
+                response_json_schema=slide_layout.json_schema,
+            ),
+        )
+        return response.model_dump(mode="json")